pdf2 0.2.0-alpha.1

A PDF 2.0 (ISO 32000-2) reader, writer and text extractor.
Documentation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
//! Content-stream interpretation, far enough to extract text (§7.8.2, §9).
//!
//! The interpreter walks a decoded content stream with the same tokenizer the
//! object parser uses: operands are pushed on a stack and the operator that
//! follows consumes them. Only the text operators change what is extracted —
//! `BT`, `ET`, `Tf`, `Td`, `TD`, `T*`, `TL`, `Tm`, `Tc`, `Tw`, `Tz`, `Ts`,
//! `Tj`, `TJ`, `'` and `"` — while everything else is consumed and ignored, so
//! graphics-heavy pages do not disturb the text run.
//!
//! Limits, on purpose and documented in the README: simple fonts only (Type1,
//! TrueType, Type3 with a single-byte encoding); composite (Type0) fonts,
//! `/ToUnicode` CMaps and `/Widths`-based gap detection are not implemented, so
//! line breaks come from text-positioning operators and spaces from the text
//! itself or from large `TJ` adjustments.

use std::collections::HashMap;

use crate::document::Document;
use crate::error::{Error, Result};
use crate::lexer::Token;
use crate::object::{Dictionary, Object};
use crate::page::Page;
use crate::parser::Parser;

/// A `TJ` adjustment larger than this (in thousandths of an em) is treated as
/// a word gap. Wire values land between 200 and 400 for a real space.
const WORD_GAP: f64 = 200.0;

/// Movement in text space smaller than this is the same line.
const LINE_TOLERANCE: f64 = 0.01;

/// A simple font: one byte per code, resolved to characters.
#[derive(Debug, Clone)]
struct Font {
    /// Code → character, index by byte value.
    encoding: Vec<Option<char>>,
}

impl Font {
    fn character(&self, code: u8) -> Option<char> {
        self.encoding.get(usize::from(code)).copied().flatten()
    }
}

/// The text state and the glyph mapping needed to read a page.
struct TextInterpreter<'a> {
    document: &'a Document,
    resources: Option<Dictionary>,
    fonts: HashMap<Vec<u8>, Font>,
    font: Option<Font>,
    leading: f64,
    /// Current text line origin, in text space.
    line_x: f64,
    line_y: f64,
    /// Y of the last emitted text, for line-break detection.
    last_y: Option<f64>,
    output: String,
}

impl<'a> TextInterpreter<'a> {
    fn new(document: &'a Document, resources: Option<Dictionary>) -> Self {
        Self {
            document,
            resources,
            fonts: HashMap::new(),
            font: None,
            leading: 0.0,
            line_x: 0.0,
            line_y: 0.0,
            last_y: None,
            output: String::new(),
        }
    }

    fn run(mut self, content: &[u8]) -> Result<String> {
        let mut parser = Parser::new(content);
        let mut operands: Vec<Object> = Vec::new();
        loop {
            match parser.peek_token()? {
                Token::Eof => break,
                Token::Keyword(_) => {
                    let keyword = match parser.next_token()? {
                        Token::Keyword(keyword) => keyword,
                        _ => return Err(Error::malformed("content-stream operator")),
                    };
                    self.execute(&keyword, &mut operands, &mut parser)?;
                    operands.clear();
                }
                _ => {
                    let object = parser.parse_object()?;
                    operands.push(object);
                }
            }
        }
        Ok(self.output)
    }

    fn execute(
        &mut self,
        operator: &[u8],
        operands: &mut [Object],
        parser: &mut Parser<'_>,
    ) -> Result<()> {
        match operator {
            b"BT" => {
                self.line_x = 0.0;
                self.line_y = 0.0;
            }
            b"ET" => {}
            b"Tf" => {
                let name = match operands.first() {
                    Some(Object::Name(name)) => name.clone(),
                    _ => return Err(Error::malformed("/Tf needs a font name")),
                };
                // The size is not needed: line breaks come from positioning
                // operators, not from advance widths.
                self.font = Some(self.load_font(&name)?);
            }
            b"Td" | b"TD" => {
                let tx = number(operands.first());
                let ty = number(operands.get(1));
                self.line_x += tx;
                self.line_y += ty;
                if operator == b"TD" {
                    self.leading = -ty;
                }
            }
            b"T*" => {
                self.line_y -= self.leading;
            }
            b"TL" => {
                self.leading = number(operands.first());
            }
            b"Tm" => {
                // Only the translation part matters for text extraction.
                self.line_x = number(operands.get(4));
                self.line_y = number(operands.get(5));
            }
            // Character spacing, word spacing, horizontal scaling and rise
            // do not change which characters a text run produces.
            b"Tc" | b"Tw" | b"Tz" | b"Ts" => {}
            b"Tj" => {
                if let Some(Object::String(string)) = operands.first() {
                    self.show(string.as_bytes());
                }
            }
            b"'" => {
                self.line_y -= self.leading;
                if let Some(Object::String(string)) = operands.first() {
                    self.show(string.as_bytes());
                }
            }
            b"\"" => {
                // Set word and character spacing, then show on the next line.
                self.line_y -= self.leading;
                if let Some(Object::String(string)) = operands.get(2) {
                    self.show(string.as_bytes());
                }
            }
            b"TJ" => {
                if let Some(Object::Array(items)) = operands.first() {
                    for item in items {
                        match item {
                            Object::String(string) => self.show(string.as_bytes()),
                            Object::Integer(adjustment) => self.adjust(*adjustment as f64),
                            Object::Real(adjustment) => self.adjust(*adjustment),
                            _ => return Err(Error::malformed("unsupported /TJ array entry")),
                        }
                    }
                }
            }
            b"BI" => {
                // Inline images are opaque data; text cannot hide inside them,
                // so skip to the `EI` that closes them.
                skip_inline_image(parser)?;
            }
            _ => {}
        }
        Ok(())
    }

    fn adjust(&mut self, adjustment: f64) {
        // Positive values move the pen left, so a wide negative adjustment is
        // how writers usually encode a space inside a `TJ` array.
        if -adjustment >= WORD_GAP && !self.output.ends_with(' ') {
            self.output.push(' ');
        }
    }

    fn show(&mut self, bytes: &[u8]) {
        if self
            .last_y
            .is_none_or(|y| (y - self.line_y).abs() > LINE_TOLERANCE)
        {
            if !self.output.is_empty() && !self.output.ends_with('\n') {
                self.output.push('\n');
            }
            self.last_y = Some(self.line_y);
        }
        let Some(font) = &self.font else {
            return;
        };
        for &code in bytes {
            if let Some(character) = font.character(code) {
                self.output.push(character);
            }
        }
    }

    fn load_font(&mut self, name: &[u8]) -> Result<Font> {
        if let Some(font) = self.fonts.get(name) {
            return Ok(font.clone());
        }
        let font = self.read_font(name)?;
        self.fonts.insert(name.to_vec(), font.clone());
        Ok(font)
    }

    fn read_font(&self, name: &[u8]) -> Result<Font> {
        let font = match self
            .resources
            .as_ref()
            .and_then(|resources| resources.get(b"Font"))
        {
            Some(fonts) => {
                let fonts = self.document.resolve(Some(fonts))?.unwrap_or(Object::Null);
                match fonts {
                    Object::Dictionary(fonts) => fonts.get(name).cloned(),
                    _ => None,
                }
            }
            None => None,
        };
        let Some(font) = font else {
            return Err(Error::Malformed {
                what: "content stream selects a font the page does not declare",
                offset: None,
            });
        };
        let font = self.document.resolve(Some(&font))?.unwrap_or(Object::Null);
        let Object::Dictionary(dict) = font else {
            return Err(Error::Malformed {
                what: "font resource is not a dictionary",
                offset: None,
            });
        };
        if let Some(Object::Name(subtype)) = dict.get(b"Subtype") {
            if subtype == b"Type0" {
                return Err(Error::Malformed {
                    what: "composite (Type0) fonts are not supported yet",
                    offset: None,
                });
            }
        }
        Ok(Font {
            encoding: self.encoding_of(&dict)?,
        })
    }

    fn encoding_of(&self, dict: &Dictionary) -> Result<Vec<Option<char>>> {
        let mut encoding = standard_encoding();
        let Some(encoding_object) = dict.get(b"Encoding") else {
            return Ok(encoding);
        };
        let encoding_object = self
            .document
            .resolve(Some(encoding_object))?
            .unwrap_or(Object::Null);
        match encoding_object {
            Object::Name(name) => encoding = base_encoding(&name),
            Object::Dictionary(differences) => {
                if let Some(Object::Name(base)) = differences.get(b"BaseEncoding") {
                    encoding = base_encoding(base);
                }
                let Some(differences) = differences.get(b"Differences") else {
                    return Ok(encoding);
                };
                let differences = self
                    .document
                    .resolve(Some(differences))?
                    .unwrap_or(Object::Null);
                let Object::Array(items) = differences else {
                    return Err(Error::Malformed {
                        what: "/Differences must be an array",
                        offset: None,
                    });
                };
                let mut code = 0_usize;
                for item in &items {
                    match item {
                        Object::Integer(value) => {
                            code = usize::try_from(*value).map_err(|_| Error::Malformed {
                                what: "/Differences code must be a non-negative integer",
                                offset: None,
                            })?;
                        }
                        Object::Name(glyph) => {
                            if code < 256 {
                                encoding[code] = glyph_to_char(glyph);
                            }
                            code += 1;
                        }
                        _ => {
                            return Err(Error::Malformed {
                                what: "/Differences entries must be integers or names",
                                offset: None,
                            })
                        }
                    }
                }
            }
            _ => {}
        }
        Ok(encoding)
    }
}

/// Skip inline image data: everything up to the `EI` operator.
fn skip_inline_image(parser: &mut Parser<'_>) -> Result<()> {
    let search = parser.raw_position();
    let found = parser
        .content()
        .get(search..)
        .and_then(|rest| find_operator(rest, b"EI"))
        .ok_or(Error::malformed("inline image without a closing EI"))?;
    parser.seek(search + found + 2);
    Ok(())
}

/// Find `needle` as a whole operator: preceded by whitespace or a delimiter.
fn find_operator(haystack: &[u8], needle: &[u8]) -> Option<usize> {
    let mut index = 0;
    while index + needle.len() <= haystack.len() {
        if &haystack[index..index + needle.len()] == needle {
            let before_ok = index == 0
                || crate::lexer::is_whitespace(haystack[index - 1])
                || crate::lexer::is_delimiter(haystack[index - 1]);
            let after = index + needle.len();
            let after_ok = after >= haystack.len()
                || crate::lexer::is_whitespace(haystack[after])
                || crate::lexer::is_delimiter(haystack[after]);
            if before_ok && after_ok {
                return Some(index);
            }
        }
        index += 1;
    }
    None
}

fn number(operand: Option<&Object>) -> f64 {
    match operand {
        Some(Object::Integer(value)) => *value as f64,
        Some(Object::Real(value)) => *value,
        _ => 0.0,
    }
}

impl Document {
    /// Extract the text of a page's content streams.
    pub fn extract_text(&self, page: &Page) -> Result<String> {
        let content = self.page_content(page)?;
        self.extract_text_from(&content, page.resources.as_ref())
    }

    /// Extract text from a decoded content stream and its resource dictionary.
    pub fn extract_text_from(
        &self,
        content: &[u8],
        resources: Option<&Dictionary>,
    ) -> Result<String> {
        TextInterpreter::new(self, resources.cloned()).run(content)
    }
}

/// The ASCII part of `StandardEncoding` (§D.2), which is what almost every
/// simple font uses for its codes 32–126.
fn standard_encoding() -> Vec<Option<char>> {
    let mut encoding = vec![None; 256];
    for code in 32_u8..=126 {
        encoding[usize::from(code)] = Some(char::from(code));
    }
    encoding[0x27] = Some('\u{2019}'); // quoteright
    encoding[0x60] = Some('\u{2018}'); // quoteleft
    encoding
}

/// `WinAnsiEncoding` is Latin-1 with the `0x80..=0x9F` range replaced by the
/// Windows-1252 punctuation block (§D.2).
fn win_ansi_encoding() -> Vec<Option<char>> {
    let mut encoding = vec![None; 256];
    for code in 0_u8..=255 {
        encoding[usize::from(code)] = match code {
            0x80 => Some('\u{20AC}'),
            0x82 => Some('\u{201A}'),
            0x83 => Some('\u{0192}'),
            0x84 => Some('\u{201E}'),
            0x85 => Some('\u{2026}'),
            0x86 => Some('\u{2020}'),
            0x87 => Some('\u{2021}'),
            0x88 => Some('\u{02C6}'),
            0x89 => Some('\u{2030}'),
            0x8A => Some('\u{0160}'),
            0x8B => Some('\u{2039}'),
            0x8C => Some('\u{0152}'),
            0x8E => Some('\u{017D}'),
            0x91 => Some('\u{2018}'),
            0x92 => Some('\u{2019}'),
            0x93 => Some('\u{201C}'),
            0x94 => Some('\u{201D}'),
            0x95 => Some('\u{2022}'),
            0x96 => Some('\u{2013}'),
            0x97 => Some('\u{2014}'),
            0x98 => Some('\u{02DC}'),
            0x99 => Some('\u{2122}'),
            0x9A => Some('\u{0161}'),
            0x9B => Some('\u{203A}'),
            0x9C => Some('\u{0153}'),
            0x9E => Some('\u{017E}'),
            0x9F => Some('\u{0178}'),
            0x81 | 0x8D | 0x8F | 0x90 | 0x9D => None,
            _ => Some(char::from(code)),
        };
    }
    encoding
}

fn base_encoding(name: &[u8]) -> Vec<Option<char>> {
    match name {
        b"WinAnsiEncoding" => win_ansi_encoding(),
        // MacRomanEncoding shares the ASCII range with StandardEncoding; the
        // upper half is not implemented, which the README states.
        b"StandardEncoding" | b"MacRomanEncoding" | b"PDFDocEncoding" => standard_encoding(),
        _ => standard_encoding(),
    }
}

/// Glyph names to characters, for `/Differences` arrays.
///
/// Covers the Latin punctuation and accented letters that real documents use;
/// anything else is skipped rather than guessed.
fn glyph_to_char(name: &[u8]) -> Option<char> {
    if name.len() == 1 {
        return Some(char::from(name[0]));
    }
    let name = std::str::from_utf8(name).ok()?;
    let named = match name {
        "space" => ' ',
        "exclam" => '!',
        "quotedbl" => '"',
        "numbersign" => '#',
        "dollar" => '$',
        "percent" => '%',
        "ampersand" => '&',
        "quotesingle" => '\'',
        "parenleft" => '(',
        "parenright" => ')',
        "asterisk" => '*',
        "plus" => '+',
        "comma" => ',',
        "hyphen" => '-',
        "period" => '.',
        "slash" => '/',
        "colon" => ':',
        "semicolon" => ';',
        "less" => '<',
        "equal" => '=',
        "greater" => '>',
        "question" => '?',
        "at" => '@',
        "bracketleft" => '[',
        "backslash" => '\\',
        "bracketright" => ']',
        "asciicircum" => '^',
        "underscore" => '_',
        "grave" => '`',
        "braceleft" => '{',
        "bar" => '|',
        "braceright" => '}',
        "asciitilde" => '~',
        "bullet" => '\u{2022}',
        "endash" => '\u{2013}',
        "emdash" => '\u{2014}',
        "quoteleft" => '\u{2018}',
        "quoteright" => '\u{2019}',
        "quotedblleft" => '\u{201C}',
        "quotedblright" => '\u{201D}',
        "ellipsis" => '\u{2026}',
        "dagger" => '\u{2020}',
        "daggerdbl" => '\u{2021}',
        "section" => '\u{00A7}',
        "paragraph" => '\u{00B6}',
        "degree" => '\u{00B0}',
        "plusminus" => '\u{00B1}',
        "multiply" => '\u{00D7}',
        "divide" => '\u{00F7}',
        "copyright" => '\u{00A9}',
        "registered" => '\u{00AE}',
        "trademark" => '\u{2122}',
        "eacute" => '\u{00E9}',
        "egrave" => '\u{00E8}',
        "agrave" => '\u{00E0}',
        "ccedilla" => '\u{00E7}',
        "udieresis" => '\u{00FC}',
        "odieresis" => '\u{00F6}',
        "adieresis" => '\u{00E4}',
        "germandbls" => '\u{00DF}',
        "fi" => '\u{FB01}',
        "fl" => '\u{FB02}',
        "zero" => '0',
        "one" => '1',
        "two" => '2',
        "three" => '3',
        "four" => '4',
        "five" => '5',
        "six" => '6',
        "seven" => '7',
        "eight" => '8',
        "nine" => '9',
        _ => {
            // `uniXXXX` and `uXXXX` names carry their code point directly.
            let hex = name
                .strip_prefix("uni")
                .or_else(|| name.strip_prefix('u'))?;
            return char::from_u32(u32::from_str_radix(hex, 16).ok()?);
        }
    };
    Some(named)
}

#[cfg(test)]
mod tests {
    use super::{find_operator, glyph_to_char, standard_encoding, win_ansi_encoding};

    #[test]
    fn win_ansi_uses_latin1_outside_the_punctuation_block() {
        let encoding = win_ansi_encoding();
        assert_eq!(encoding[usize::from(b'A')], Some('A'));
        assert_eq!(encoding[0xE9], Some('\u{00E9}'), "e-acute");
        assert_eq!(encoding[0x80], Some('\u{20AC}'), "euro");
        assert_eq!(encoding[0x95], Some('\u{2022}'), "bullet");
        assert_eq!(encoding[0x81], None, "undefined in cp1252");
    }

    #[test]
    fn standard_encoding_is_ascii_with_the_two_quote_names_swapped() {
        let encoding = standard_encoding();
        assert_eq!(encoding[usize::from(b' ')], Some(' '));
        assert_eq!(encoding[usize::from(b'~')], Some('~'));
        assert_eq!(encoding[0x27], Some('\u{2019}'));
        assert_eq!(encoding[0x60], Some('\u{2018}'));
        assert_eq!(encoding[0x00], None);
    }

    #[test]
    fn glyph_names_resolve_to_characters() {
        assert_eq!(glyph_to_char(b"A"), Some('A'));
        assert_eq!(glyph_to_char(b"7"), Some('7'));
        assert_eq!(glyph_to_char(b"bullet"), Some('\u{2022}'));
        assert_eq!(glyph_to_char(b"eacute"), Some('\u{00E9}'));
        assert_eq!(glyph_to_char(b"uni20AC"), Some('\u{20AC}'));
        assert_eq!(glyph_to_char(b"u0041"), Some('A'));
        assert_eq!(glyph_to_char(b"nosuchglyph"), None);
    }

    #[test]
    fn inline_image_terminator_must_be_a_whole_operator() {
        assert_eq!(find_operator(b"abc EI more", b"EI"), Some(4));
        assert_eq!(find_operator(b"EI", b"EI"), Some(0));
        assert_eq!(
            find_operator(b"tailEIx", b"EI"),
            None,
            "EI inside a word is not the operator"
        );
        assert_eq!(find_operator(b"no terminator", b"EI"), None);
    }
}